Abstract:We develop a deep convolutional neural network (DCNN) based framework for model-free prediction of the occurrence of extreme events both in time ("when") and in space ("where") in nonlinear physical systems of spatial dimension two. The measurements or data are a set of two-dimensional snapshots or images. For a desired time horizon of prediction, a proper labeling scheme can be designated to enable successful training of the DCNN and subsequent prediction of extreme events in time. Given that an extreme event has been predicted to occur within the time horizon, a space-based labeling scheme can be applied to predict, within certain resolution, the location at which the event will occur. We use synthetic data from the 2D complex Ginzburg-Landau equation and empirical wind speed data of the North Atlantic ocean to demonstrate and validate our machine-learning based prediction framework. The trade-offs among the prediction horizon, spatial resolution, and accuracy are illustrated, and the detrimental effect of spatially biased occurrence of extreme event on prediction accuracy is discussed. The deep learning framework is viable for predicting extreme events in the real world.

What problem does this paper attempt to address?

### Problems the paper attempts to solve The paper aims to solve the **problem of spatio - temporal prediction of extreme events in nonlinear physical systems**. Specifically, the author has developed a framework based on deep convolutional neural network (DCNN) to predict the time and location of extreme events in two - dimensional space. The following are the core problems of the paper: 1. **When (When):** - **Problem description**: How to predict whether an extreme event will occur within a certain future time interval. - **Method**: By dividing the system state (image) into two categories for labeling: one is the state within the time interval before the occurrence of the extreme event (labeled as 1), and the other is the state without extreme events (labeled as 0). Then use these labeled images to train the DCNN model to predict whether an extreme event will occur within a certain future time interval. 2. **Where (Where):** - **Problem description**: If it is predicted that an extreme event will occur in the future, how to further predict the specific location where it will occur. - **Method**: Divide the spatial area into grids and assign a label to each grid cell. For example, for a \(4\times4\) grid, each cell can be labeled from 0 to 15. If an extreme event occurs within a certain cell, the corresponding image will be labeled with the label of that cell. In this way, the DCNN can be trained to predict the specific location of the extreme event. ### Research background and motivation Extreme events (such as earthquakes, strong tropical cyclones, tornadoes and abnormal waves in the ocean) occur frequently in nature and engineering systems, causing serious damage to society, environment and economy. However, due to the lack of accurate dynamic models, the existing model - based prediction methods face challenges in practical applications. Therefore, it is particularly important to develop a data - based model - free prediction method. ### Main contributions 1. **Model - free prediction framework**: A model - free prediction framework based on DCNN is proposed, which can predict the time and location of extreme events only relying on data. 2. **Spatio - temporal prediction**: It not only solves the problem of "when" it occurs, but also solves the problem of "where" it occurs, providing more comprehensive prediction capabilities. 3. **Experimental verification**: Verification is carried out using synthetic data (from the 2D complex Ginzburg - Landau equation) and real - world atmospheric system data (wind speed distribution in the Arctic Ocean), demonstrating the effectiveness and robustness of this method. ### Trade - offs of key indicators The paper discusses the trade - off relationships among three key indicators: - **Prediction Horizon** - **Spatial Resolution** - **Prediction Accuracy** Generally speaking, increasing the prediction time range or spatial resolution will reduce the prediction accuracy. Therefore, choosing an appropriate combination of parameters is the key to achieving effective prediction. ### Summary By introducing the DCNN - based framework, the paper successfully solves the problem of spatio - temporal prediction of extreme events in nonlinear physical systems. This method not only has theoretical significance, but also shows potential value in practical applications, especially in dealing with extreme events in natural disasters and other complex systems.

Predicting extreme events from data using deep machine learning: when and where

Spatio-temporal probability prediction of rainfall-induced landslides based on deep learning under climate change

Survey on the Application of Deep Learning in Extreme Weather Prediction

Modeling Extreme Events in Time Series Prediction

A Novel Framework for Spatio-Temporal Prediction of Environmental Data Using Deep Learning

Deep Learning Approaches in Predicting Tropical Cyclone Tracks: an Analysis Focused on the Northwest Pacific Region

Using spatial extreme-value theory with machine learning to model and understand spatially compounding weather extremes

Short-term precipitation prediction using deep learning

Deep Learning-based Extreme Heatwave Forecast

Application of Deep Convolutional Neural Networks for Detecting Extreme Weather in Climate Datasets

Prediction of Occurrence of Extreme Events using Machine Learning

Advances and prospects of deep learning for medium-range extreme weather forecasting

Data-assisted reduced-order modeling of extreme events in complex dynamical systems

Predicting observed and hidden extreme events in complex nonlinear dynamical systems with partial observations and short training time series

Spatio temporal hydrological extreme forecasting framework using LSTM deep learning model

Predicting ice flow using machine learning

Probabilistic forecasts of extreme heatwaves using convolutional neural networks in a regime of lack of data

Skillful Extended-Range Forecast of Rainfall and Extreme Events in East China Based on Deep Learning

Model-free prediction of emergence of extreme events in a parametrically driven nonlinear dynamical system by Deep Learning

Deep learning rainfall-runoff predictions of extreme events

Inductive Predictions of Extreme Hydrologic Events in The Wabash River Watershed